Towards a Complete Digital Twin for Photovoltaic Systems: A Comparative Study of Prediction Approaches

Abstract

Photovoltaic installations often underperform relative to expectations, resulting in a gap between theoretical and actual production that poses a major challenge for the energy transition. To address this issue, we propose a modular and unified Digital Twin (DT) architecture, assigning a dedicated DT to each component (e.g., panels and inverters). This study focuses on the first pillar of the framework: the prediction of photovoltaic power. Using a high-resolution data set spanning an entire year and accounting for seasonal variations, we conducted a comparative study of Machine Learning (ML) and Deep Learning (DL) models to identify the most effective and suitable approach for integration into the Digital Twin.